Transforming Indigenous Malaria Knowledge into Digital Health Intelligence Using Retrieval-Augmented Language Models

Indigenous malaria knowledge plays an important role in community health decision-making across many regions of Africa, yet it remains largely absent from digital health intelligence systems. This study investigates whether culturally embedded malaria knowledge can be computationally represented and operationalised using artificial intelligence. An indigenous malaria knowledge corpus was constructed from ethnobotanical studies, community health literature, and documented cultural narratives across 11 Ugandan communities, including 38 medicinal plant profiles, 17 non-plant prevention practices, and 14 explanatory belief pathways. The knowledge was transformed into 812 prompt-response training pairs and used to fine-tune a GPT-NeoX-1.3B language model using Low-Rank Adaptation (LoRA) combined with retrieval-augmented knowledge grounding. Experimental evaluation demonstrated strong performance across several dimensions. The model achieved response accuracy of 0.86, with retrieval performance reaching precision@3 of 0.81 and recall@5 of 0.77. Human expert assessment further indicated high cultural authenticity (mean score 4.3/5) and strong safety compliance (0.94) for danger-sign escalation scenarios. Retrieval grounding significantly improved reliability, reducing hallucination rates from 28% in the base model to 9% in the fine-tuned system. Cross-community evaluation showed moderate knowledge transfer (0.69), with prevention practices generalising more effectively than district-specific herbal treatments. In addition, computational testing demonstrated that the system can operate within low-resource environments, producing responses in under three seconds using CPU-based inference. These findings demonstrate the feasibility of transforming indigenous malaria knowledge into interpretable digital health intelligence while preserving cultural reasoning and safety considerations. The study provides a proof-of-concept framework for integrating indigenous knowledge systems into artificial intelligence-enabled health technologies in low-resource settings.